Bartek Ogryczak

• Advising an early-stage venture on applying large language models in the legal domain.
• Architecture built around local and self-hosted inference, keeping confidential material inside the client's own environment.
• System design across model selection, evaluation, and data handling.

Details are limited while the company is in stealth.

• Cloud Inference Team: developed, operated, and scaled Claude inference infrastructure on Google Vertex AI. Scaled from near zero to one of Anthropic's largest enterprise inference channels.
• DRI for 6+ Claude model launches on Vertex AI.
• DRI for bringing Claude capabilities — including Thinking, Prompt Caching, Token Counting, and Multimodal Inputs — to Google Vertex AI.
• Core contributor to Dystro, Anthropic's inference coordination layer responsible for routing traffic across thousands of accelerators supporting large-scale Claude inference.
• Implemented key components of Vertex's Provisioned Throughput feature for Claude models.

• Errors and Issues Team: a cross-functional team responsible for building and maintaining Sentry's core product.
• Automated enforcement of logging best practices, preventing outages due to misconfiguration.
• Improved error grouping accuracy and maintainability within Sentry's core error-processing pipeline.
• Implemented caching to reduce ClickHouse data scans by up to 97%.
• Partnered with ML engineers to prototype LLM-based approaches for automated error analysis and triage.

• Application Foundry Team: developed tools to enhance service resilience and scalability. Managed both application code and infrastructure setup.
• AWS Migration: key role in a cross-functional “tiger” team that migrated infrastructure company-wide from RackSpace to AWS EKS.
• Standardized task scheduling by migrating to Kubernetes CronJobs.
• Developed Hawker, an in-house messaging system, improving cross-service communication.
• Implemented Twirp for standardized RPC, enhancing team interoperability.

• Back-end Foundry Squad / Infrastructure Team: led architectural solutions and system implementations to enable scalable operations.
• Designed and implemented Cerberus, an advanced permissions system.
• Co-created PySOA, enabling scalable service-oriented architecture and reducing remote call latency.

Designed and implemented scalable architecture for Layar's augmented reality applications.

• Hyves (Amsterdam) — back-end engineer at the Netherlands' largest social network; managed 2 billion image/video files with a replication algorithm that reduced latency and increased reliability.
• Red Trabajar (Madrid) — led a rewrite of Spain's largest job portal (Zend Framework, Oracle); built a system sending 15 million customized emails nightly.
• Santander Bank (Madrid) — real-time data integration for investment-banking risk management; devised an algorithm, running on a computing cluster, that cut latency and runtime 12-fold.
• Melog.com (Warsaw) — Linux sysadmin and back-end specialist on Oferty.net, a real-estate meta-search platform.

Specialized in Distributed Systems; developed USOSweb for the thesis — a component of the award-winning USOS system now used by over 50 universities.